CIAG: Conditional Idempotent Association Generation for Heterogeneous Track-to-Track Association

Abstract

In advanced defence and security systems, multi-sensor fusion is widely used to improve the overall observation capability, and heterogeneous sensors are a typical deployment in multi-sensor systems. Track-to-track association (T2TA) of heterogeneous sensors is the precondition and foundation of heterogeneous sensor track fusion. However, problems such as ubiquitous systematic and random errors, inconsistent update periods and features caused by two heterogeneous sensors bring significant challenges to T2TA and existing methods have not solved the above problems adequately. To address these problems, we propose conditional idempotent association generation for heterogeneous track-to-track association (CIAG). In CIAG, a track state mapping module (TSMM) is constructed to unify asynchronous and heterogeneous tracks from heterogeneous sensors. The TSMM can also mitigate the effects of systematic and random errors. An idempotent association generation module (IAGM) is constructed to model tracks and association matrices jointly, and generate association matrices directly and precisely. Moreover, CIAG realises an end-to-end generation from the track tensor to the association matrix that can avoid long time consumption caused by traversal calculations of tracks. Comprehensive experiments demonstrate that CIAG can achieve the best association performance and has better association efficiency.